1: \begin{abstract}

2: We study min-max algorithms to solve zero-sum differentiable games on 

3: Riemannian manifold. 

4: The notions of 

5: differentiable Stackelberg equilibrium and 

6: differentiable Nash equilibrium in Euclidean space

7: are generalized to Riemannian manifold,

8: through an intrinsic definition which does not depend 

9: on the choice of local coordinate chart of manifold. 

10: We then provide sufficient conditions for the local convergence

11: of the deterministic simultaneous algorithms $\tau$-GDA and $\tau$-SGA

12: near such equilibrium, using a general methodology 

13: based on spectral analysis. 

14: These algorithms are extended with stochastic gradients 

15: and applied to the training of Wasserstein GAN.

16: The discriminator of GAN is constructed from 

17: Lipschitz-continuous functions based on Stiefel manifold.

18: We show numerically how the insights obtained from the 

19: local convergence analysis may lead to 

20: an improvement of GAN models.

21: \end{abstract}

22: